[GHC] #13344: Core string literal patch regresses compiler performance considerably

[GHC]

#13344: Core string literal patch regresses compiler performance considerably
-------------------------------------+-------------------------------------
        Reporter:  bgamari           |                Owner:  bgamari
            Type:  bug               |               Status:  new
        Priority:  high              |            Milestone:  8.2.1
       Component:  Compiler          |              Version:  8.1
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Compile-time      |  Unknown/Multiple
  performance bug                    |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by bgamari):

 Just looking at the interface file sizes between the two trees, it appears
 that many files increased by a few hundred bytes with a few (e.g.
 `GHC.Show`, `GHC.Generics`) increasing by some kilobytes. For future
 reference I performed the comparison with,
 {{{#!sh
 echo "delta after before file"
 paste \
     <(ls -l $(find ghc-compare-1/libraries/base/dist-install -iname *.hi)
 | awk '{print $5,$9}') \
     <(ls -l $(find ghc-compare-2/libraries/base/dist-install -iname *.hi)
 | awk '{print $5,$9}') \
     | awk '{ print (($1 - $3)), $1, $3, $2}'
     | sort -rn | less
 }}}

 Which results in
 {{{
 16461 800034 783573 libraries/base/dist-install/build/GHC/Generics.hi
 7673 130677 123004 libraries/base/dist-install/build/GHC/RTS/Flags.hi
 7555 969695 962140 libraries/base/dist-install/build/Data/Data.hi
 7449 103629 96180 libraries/base/dist-install/build/GHC/IO/Exception.hi
 7068 268944 261876 libraries/base/dist-install/build/Data/Foldable.hi
 6904 136200 129296 libraries/base/dist-install/build/GHC/Show.hi
 4830 211720 206890 libraries/base/dist-install/build/Data/Monoid.hi
 4767 458028 453261 libraries/base/dist-install/build/Data/Semigroup.hi
 4735 498885 494150 libraries/base/dist-install/build/Foreign/C/Types.hi
 4651 59316 54665 libraries/base/dist-install/build/GHC/IO/Handle/Types.hi
 ...
 }}}

 In the case of `GHC.Generics` we now have things like,
 {{{#!hs
   $fEqV14 :: [Char]
   {- Unfolding: (unpackCString# $fEqV15) -}

   $fEqV15 :: Addr#
   {- HasNoCafRefs, Unfolding: ("error"#) -}
 }}}

 Whereas we previously had,
 {{{#!hs
   $fEqV11 :: [Char]
   {- Unfolding: (unpackCString# "error"#) -}
 }}}

 One potential improvement we can make here (at the expense of some
 constant folding opportunities) is to drop the unfoldings from `Addr#`
 bindings; afterall, in most cases there is little benefit to inlining the
 string literal.

 In the case of the `Typeable` bindings, I'm very tempted to just eliminate
 unfoldings from all generated `TrName`s. This would likely cut down on
 interface file sizes dramatically.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13344#comment:10>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler